1.1 Background
The question about plants and animals’ current distribution is discussed for a long history and makes many ecologists find the explanation. Many modelers root in species-environment relationships to establish many modeling approaches for solving this question (Guisan and Thuiller, 2005). Analysis of the species’ geographic distribution has always been an important issue in vegetation science, and is currently focused by other sub-disciplines such as biogeography and landscape ecology. The relationship between environmental gradients and vegetation distribution is one of the most important issues examined in vegetation science (Miller et al., 2007). The ability to quantify the relationship leads to predict potential distribution of vegetation and is applicable for predicting spatial distribution under changing environmental conditions, such as climate change occurring (Miller et al., 2007).
The purpose of potential vegetation field survey is to understand plant ecology and apply to ecological conservation, landscape restoration, and landscape planting (Yang, 1997). Survey data for estimating or predicting potential vegetation according to plant ecology, or plant geography, explain forest composition, structure, and function, further more, the relationship with environmental variables and its succession stage. Those data information allow us to predict the next succession, current or future distribution of species and vegetation and are useful for forest ecosystem management, biological conservation, and landscape restoration. On the contrary to traditional survey analysis,
2
Chiou et al. (2006) introduced GIS technique and several models are rapidly developed in recent years. Combining vegetation mapping and analysis of satellite images with GIS generate predictive vegetation model, a new approach for analysis species/vegetation and environmental variables (such as Gu et al., 2006; Tsao, 2007;
Yen, 2007) (Franklin, 1995; Guisan and Zimmermann, 2000; Scott et al., 2002) and this study is also based on the new approach.
Climate change has become an important focused issue in recent years, as a basis for assessing whether anthropogenic greenhouse effect has enhanced climate change and how the continuingly growing greenhouse gas concentration will lead to an unknown future climate. According to the IPCC’s Third Assessment Reports (TAR, IPCC, 2001), the average temperature of global surface has increased by about 0.6 °C in the past century, and to the IPCC’s Forth Assessment Report (AR4), warming in the last 100 years has increased by 0.74 °C in global average temperature. This is above the 0.6
°C increase in the 20th century prior to the Third Assessment Report (IPCC, 2007).
Taiwan has been moving toward a warmer and drier climate. Enhanced precipitation is observed in the limited areas in the limited times, however, a systematic trend (or change) is not observed (Hsu, 2002). In general, under constant climate change and global warming conditions possibly forces the distribution area of current vegetation diminishing, and increases the risk of species extinction. To predict the change of distribution of species under different climate scenarios is essential to assess the risk of species extinction under climate change (Thomas et al., 2004). Thus, the first mission is to establish predictable models for species potential distribution range (Tsao, 2007).
3
Previous ecologist in Taiwan mainly focused on classifying the plant communities and identifying the relationships between plant societies and environmental variables (i.e. Su, 1984a; 1984b). Island of Taiwan has a great diversity of fauna and flora due to a high degree of topographical complexity and an about 4000 m variation in altitude (elevation). A large proportion of Taiwan is not easy to access for field survey due to its hilly topography. Available field data might not be completely enough to support decision making of conservational or environmental policies (Song et al., 2007).
Species distribution models provide a possible way to fill up the gap of incomplete vegetation data (Franklin, 1998).
Projections of species distribution under climate and environmental change are of great scientific and social relevance, and basing on species distribution models (SDMs) make some assumptions such as species not adopt to global dispersal in evolution and consistency of limiting factors (Dormann, 2007). Although some of the assumptions are ecologically untenable, the predictions of the SDMs are still a useful reference to policy maker for climate change impact assessment and conservation management. This study examines the relationships between distributions of dominant species, Taiwan Hemlock (Tsuga chinensis (Franch.) Pritz. ex Diels var. formosana (Hayata) Li and Keng) of alpine forest in Taiwan, and climatic and topographical environmental variables. Not only traditional statistical approaches, but importance of new direction of data mining approaches in analyzing relationships between species and environmental factors will lead more precise insight and performance on SDMs prediction.
4
1.2 Objective
Although SDM are widely spread in many fields, the analysis of relationship in species-environment is still methodologically not well organized. This study will focus on comparing statistical and data mining methods to find the suitable environmental layers for building the spatial distribution of Taiwan Hemlock vegetation. Additionally, some studies of SDM application can be found in Taiwan (Gu et al., 2006; Tsao, 2007;
Yen, 2007) but fewer studies in Taiwan considers the comparison of difference model setting (like Song, 2007). Smaller grid size of the predicted background can reflect the more detail of topographical variables than climatic variables but is time consuming due to a very large data size for model estimation. Contrarily, larger grid size of background is much faster when calculating but lacks of or reduces detail information of the meso-environmental variables. Thus, this study also takes grid size, predicted area, locality units, and environmental selection of the model input into account.
Pervious studies for SDMs (mentioned later in Ch. 2 literature review) were using multiple model combination to improve the predictive accuracy for each species’ spatial distribution, however, combining different hierarchy units of species and vegetation to increase the predictive accuracy are seldom seen in resent researches. Therefore, comparing model techniques combination and species-vegetation unit combination is to compare and combine different SDMs to synthesize potential vegetation map of Taiwan Hemlock another goal of this study. The combination criteria are utilized to determine the binomial potential distribution area that Taiwan Hemlock may occur. There are 4 main objectives listed as follow:
5
1. Using data mining approach (classification and regression tree, CART and conditional inference tree, CIT) compared to statistical approaches (detrened correspondence analysis, DCA, principal component analysis, PCA and correlation analysis, CA) for analysis of relationship between species distributions and environmental variables.
2. Assess how localities inputs of different vegetation and species based sub-units on distribution modeling relate to environmental variables.
3. Evaluate how difference grid resolution affects the model performance and relationship between target species and environmental variables.
4. Model combination and synthesis of potential nature vegetation maps.
6